ANNUAL RANGE OF TEMPERATURE AND INTRA-SAMPLE TIME AVERAGING IN HIGH-RESOLUTION GEOCHEMICAL RECORDS OF TEMPERATURE SEASONALITY

The annual range of temperature (ART) is a basic aspect of present and past climates. One approach to estimating past ART uses isotopic or elemental ratios in skeletal hard parts that have an accretionary mode of growth. Even the highest resolution sampling schemes still result in the physical homogenization of skeletal material representing days to weeks of growth. This intra-sample time averaging necessarily reduces the observable ART. We report a study of daily temperature records that suggests a means of reducing the effects of intra-sample time averaging in proxy records of seasonality. Based on daily temperature records from weather stations, we demonstrate that ART is a function of the number of days averaged to produce the maximum and minimum temperatures from which ART is calculated. Calculating ART from the maximum and minimum daily temperatures yields the maximum ART. Averaging successive days in non-overlapping windows (an analog for time averaging of skeletal samples) results in a logarithmic decrease in ART with increasing window size (N) for N up to about 90 days. For records from each of 137 U.S. weather stations within 0.5° longitudinally of 100° W, we calculated mean ART(N) and least squares linear regressions of ART(N) on log(N) as for N?90. Mean r² for the individual station regressions was 0.998 and the average offset between the intercepts and values of ART(1) was 0.91° F. Using synthetic temperature records and a quantitative model, we show that the dependence of ART(N) on log(N) is a function of short-term autocorrelations in the variance structure of daily temperatures that are superimposed on seasonal variation in temperature. The linear relationship between ART(N) and log(N) is the key to estimating ART(N?90) from the geochemistry of time averaged samples. Serial samples with known temporal resolution provide a single value of ART. Coarsening the resolution by averaging measured values from successive, non-overlapping sets of samples provides additional estimates of ART at higher N. Given the strong linear relationship between ART(N) and log(N) for temperature records, only a few values of ART(N) based on the geochemical proxy are necessary to calculate a linear regression and estimate any value of ART for N?90, effectively eliminating the effects of intra-sample time averaging.